Hong Kong Insilico Medicine has published a study showing that their deep learning system can identify potential treatments for fibrosis. This system, called the generative tensor training system with reinforcement or, briefly, GENTRL, was able to detect six promising therapies in just 21 days. One of these methods has shown promising results in experiments on experimental mice. The study was published in the journal Nature Biotechnology, and the source code for the model was available on Github.
“We have the strategic thinking of Artificial Intelligence in combination with his imagination”says Insilico CEO Alexander Zhavoronkov, who compares GENTRL's work with AlphaGo's machine learning system, developed by Google's Deepmind, to challenge go championship players.
Zhavoronkov founded the company in 2014. He received his initial education in computer science and spent several years working at ATI, until in 2006 AMD bought the company. At this point, he decided to change his occupation and engage in biotechnological research, becoming interested in the area of slowing down the aging process. He received a master's degree from Johns Hopkins University, and then a Ph.D. from Moscow State University, where he focused on research on the use of machine learning to study the physics of molecular interactions in biological systems. He worked for several companies, but then returned to Baltimore to found Insilico.
The company's initial philosophy was to use deep learning in order to teach neural networks to bypass large libraries of molecules and find targets for exposure to drugs. However, shortly after the founding of the company, Zhavoronkov became interested in the work of Jan Goodfellow in the field of machine learning and decided to change course.
“Can we get a machine to create new molecules with new properties, instead of just checking out giant supplier libraries?” – He asked such a question. The discovery of new drugs has traditionally been through molecular screening, but the question has been whether this process can be optimized and accelerated using machine learning.
The first study based on this idea, published by the company in 2016, helped attract investment for development at the intersection of the fields of biotechnology and artificial intelligence. According to Pitchbook, Zhavoronkov raised another $ 24.3 million in investments from sponsors such as A-Level Capital and Juvenescence, with a total valuation of $ 56 million. He also has several biotechnology partners, including A2A Pharmaceuticals and TARA Biosystems.
This study is about the challenge that the company and its colleagues from the world of chemistry have thrown themselves. They asked Insilico to use their system to develop potential drugs that could interfere with the activity of domainoid receptor 1 discoidine (DDR1). DDR1 is an enzyme involved in fibrosis, and although it is not yet clear whether it regulates these processes, inhibition of its activity is considered as a possible therapy. This task formed the basis of recently published studies by a group of Genentech experts who took about 8 years to identify promising DDR1 kinase inhibitors.
General view of the Insilico research process
Insilico used GENTRL to develop new potential drugs that were subsequently synthesized, and one of them was even successfully tested in mice. Designing an artificial intelligence system took about 21 days, and the total amount of time for development, synthesis and validation took about 46 days. Although none of the drugs developed by GENTRL proved to be more effective than the inhibitors discovered by the traditional research method, the traditional method took more than 8 years and millions of dollars compared to several weeks and an approximate cost of $ 150,000.
“Their molecules are amazing, they are much better than the results of our artificial intelligence”– says Zhavoronkov. “But then again, years of work play a role here against people who are not so good at chemistry, but already do such things.”
Of course, against the background of the whole development of medicines, this is only the first step. Although it is an important milestone to demonstrate the potential of AI in identifying potential drugs, it will take years of clinical trials and millions of dollars to research before any potential drug is approved as a treatment for a particular disease.
Zhavoronkov also says that Insilico still has a lot of work to do. For him, this study is considered an important breakthrough, since it shows the prospects of using AI in the creation of drugs.
“I believe this study will reduce skepticism in global pharmaceuticals.”He says.